A method is developed for using wavenumber spectra of scalar contaminants in a boundary layer dominated by buoyancy-driven convection to estimate the magnitude of vertical turbulent fluxes. The technique is analogous to the conventional inertial-dissipation method (IDM) for obtaining fluxes from variance spectral density levels in the inertial subrange but differs in the way that turbulence scales are combined to provide an eddy density diffusivity for relating flux magnitude to variance dissipation rate. The method is illustrated using spectra and direct flux covariance data from the oceanic boundary layer at the edge of a freezing lead during the 1992 Lead Experiment in the Arctic Ocean. There, density was controlled almost exclusively by salinity, and it is argued that the turbulent length and velocity scales governing vertical exchange were the inverse of the wavenumber at the peak in the weighted salinity spectrum (considerably less than the mixed layer depth) and the cube root of the product of turbulent length scale and the buoyancy flux magnitude, respectively. The sign of the skewness of temperature or salinity time series is shown to be a robust indicator of the (negative) direction of vertical flux. The free convection approach is valid (i.e., should be used instead of the conventional IDM) only if the convective turbulent scale velocity is appreciably larger than friction velocity (square root of the Reynolds stress), so that turbulent kinetic energy dissipation is approximately equal to buoyancy production. ¿ 1998 American Geophysical Union